Calibration

V* → RealWorldQA

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

RealWorldQA is estimated from V* with a Hill curve fitted on 5 models measured on both: y = 0.0000 + (0.8530 − 0.0000)·x^6.00 / (0.45440^6.00 + x^6.00), R² = 0.89, cross-validated error 2.9 pp. It is used for 1 estimate.

Estimated model	V*	RealWorldQA	Source
Step 3.7 Flash	95.3%	84.3%	estimated ± 2.9 pp, medium confidence